Re: [PATCH v3 0/2] iommu/iova: Make the rcache depot properly flexible

[Catalin Marinas <catalin.marinas@arm.com>]

On Mon, Oct 28, 2024 at 10:04:10AM +0200, Ido Schimmel wrote:
> On Mon, Jan 15, 2024 at 09:17:04AM +0200, Ido Schimmel wrote:
> > On Fri, Jan 12, 2024 at 05:31:15PM +0200, Ido Schimmel wrote:
> > > On Thu, Jan 11, 2024 at 10:13:01AM +0000, Catalin Marinas wrote:
> > > > On Thu, Jan 11, 2024 at 10:20:27AM +0200, Ido Schimmel wrote:
> > > > > On Wed, Jan 10, 2024 at 05:58:15PM +0000, Catalin Marinas wrote:
> > > > > > Transient false positives are possible, especially as the code doesn't
> > > > > > use a double-linked list (for the latter, kmemleak does checksumming and
> > > > > > detects the prev/next change, defers the reporting until the object
> > > > > > becomes stable). That said, if a new scan is forced (echo scan >
> > > > > > /sys/kernel/debug/kmemleak), are the same objects still listed as leaks?
> > > > > > If yes, they may not be transient.
> > > > > 
> > > > > We are doing "scan" and "clear" after each test. I will disable the
> > > > > "clear" and see if the leaks persist.
> > > > 
> > > > If it is indeed a false positive
> > > 
> > > Looks like the leaks are transient. After removing the "clear" step the
> > > leaks do not seem to persist.
> > > 
> > > > you can try the patch below (I haven't given it any run-time test,
> > > > only compiled):
> > > 
> > > Will try and let you know next week.
> > 
> > Looks good. Feel free to add:
> > 
> > Tested-by: Ido Schimmel <idosch@nvidia.com>
> 
> Catalin, we have been using this patch since you posted it and haven't
> seen any issues. Can please submit it upstream?

Oh, I completely forgot about it. I'll try to do it this week. Thanks
for the reminder.

-- 
Catalin